Visual homing: a purely appearance-based approach

This paper presents an algorithm that uses visual input to perform homing for an autonomous mobile robot. An image captured at the target pose (position and orientation) is compared with the currently viewed image to determine the parameters of the next move of the robot toward the target (rotation and translation). The visual data is captured using an omnidirectional camera and images are compared using the Manhattan distance function to determine both the translation and rotation angle. Results, limitations and successes are presented and discussed.